Re: [Bug-gsl] error in multifit/multiwlinear.c in gsl-2.3 and gsl-2.4

[Patrick Alken]

Hello,

  I responded to this yesterday. The issue was the
gsl_multifit_wlinear_svd function is deprecated (replaced by
gsl_multifit_wlinear_tsvd), and I hadn't tested the old function. The
issue should be fixed now in the git repository. Could you checkout the
latest git code and test your program?

Thanks,
Patrick

On 02/11/2018 12:44 PM, Vlad Koli wrote:
> Hi,
>
> There is an error in file multifit/multiwlinear.c. It is present at least  
> in sources for gsl-2.3, gsl-2.4. 
>
>   Line 69 should be replaced (I think) from:
>
>  else if (X->size1 != work->n || X->size2 != work->p)
>
>    to
>
>  else if (X->size1 != work->nmax || X->size2 != work->pmax)  // vk:
>
> Values of "work->n", "work->p" for this file, for example, are nulls.
>
> After this replacement, re-compilation, and re-installation the tests 
> began to work correctly.
>
>
> Check, for example the test file "linfit-pseudoinv.c" quoted below.  
>
> ** compilation:
>
>  cc -o a -lm -lgsl -lgslcblas linfit-pseudoinv.c 
>
> **  execution: 
>
>  ./a
>
> ** result:
>
>> gsl: multiwlinear.c:73: ERROR: size of workspace does not match size of 
>> observation matrix
>> Default GSL error handler invoked.
>> Aborted
>
> Regards,
> Vladimir Koliadin
>
>
> ----- begin file: linfit-pseudoinv.c:---------
> // compile:
> //     cc -o a -lm -lgsl -lgslcblas linfit-pseudoinv.c 
> #include <stdio.h>
> #include <gsl/gsl_multifit.h>
>
>
> double tol = 1.e-08; // tolerance
>
> main()
> {
>      
>     int n = 3; // observations
>     int m = 2; // parameters 
>
>     size_t rank;   // rank
>     double chisq;  // sum of squares of residuals
>
>     // allocate:
>     gsl_vector *x = gsl_vector_alloc(m); // solution 
>     gsl_vector *y = gsl_vector_alloc(n); // data 
>     gsl_vector *w = gsl_vector_alloc(n); // weights
>
>     gsl_matrix *A = gsl_matrix_alloc(n, m);
>     gsl_matrix *cov = gsl_matrix_alloc(m, m); // covariances
>
>     gsl_multifit_linear_workspace* wsp = gsl_multifit_linear_alloc (n, m);
>
>
>     // set matrix
>     gsl_matrix_set(A, 0,0, 1.);
>     gsl_matrix_set(A, 0,1, 1.);
>     gsl_matrix_set(A, 1,0, 2.);
>     gsl_matrix_set(A, 1,1, 2.);
>     gsl_matrix_set(A, 2,0, 3.);
>     gsl_matrix_set(A, 2,1, 3.);
>
>     // set data 
>     gsl_vector_set(y, 0, 1.);
>     gsl_vector_set(y, 1, 2.);
>     gsl_vector_set(y, 2, 3.1);
>
>     // set weights
>     gsl_vector_set_all(w,1.);
>     gsl_vector_set(w, 0, 0.);
>     gsl_vector_set(w, 2, 100.);
>
>
>     gsl_multifit_wlinear_svd (A, w, y,  tol, &rank, x, cov, &chisq, wsp);
>  
>
>     printf("solution: x=%.4f  y=%.4f   (rank=%d, chi2=%.3f)\n", 
>             gsl_vector_get(x,0), gsl_vector_get(x,1), rank, chisq);
>
>
> }
> ----- end file: linfit-pseudoinv.c:---------
>
>
>